I’ve reviewed about fifteen AI strategies in the past six months. Most of them are rubbish.

They start with the technology. “We need to implement LLMs” or “We should deploy AI agents.” That’s backwards. You don’t need AI. You need to solve business problems that might benefit from AI.

Here’s how I actually approach AI adoption when working with enterprise IT teams.

Start with Problems, Not Solutions

The first thing I do is sit down with business units and ask what’s slowing them down. Not “where could we use AI” but “what takes too long” or “what requires too many people doing manual work.”

Last month I worked with a logistics company. Their biggest pain point wasn’t something obvious like route optimization (which everyone assumes). It was contract review. They had legal teams spending days reviewing supplier agreements that followed standard templates.

That’s a good AI use case. It’s repetitive, it’s document-based, it has clear success criteria. We didn’t need to build anything fancy. A well-configured AI document analysis tool cut review time by 60%.

Run Small Tests First

I don’t care what vendors tell you about their “enterprise-ready” solutions. Test small before you commit.

We pick one process, one team, one month. We measure before and after. If it doesn’t show clear value, we stop. If it does, we expand carefully.

The logistics company I mentioned? We started with one legal team member reviewing ten contracts. Not the whole department. Not a six-month rollout. One person, ten contracts, two weeks.

When that worked, we expanded to three people. Then the whole team. Then we looked at other document types. But we earned each expansion with results.

Address Data Quality Early

Most enterprises have terrible data quality. AI doesn’t fix that. It exposes it.

I was working with a manufacturing client who wanted to predict equipment failures. Great idea. Except their maintenance logs were a mess. Techs would write “fixed the thing” or “replaced part” with no standardization.

We spent three months cleaning and structuring data before we even started the AI pilot. That wasn’t wasted time. It was necessary time. And honestly, the structured data alone improved their maintenance process before AI entered the picture.

If your data’s not ready, your AI project will fail. I tell every client: budget as much time for data work as you do for AI implementation. Usually more.

Build Internal Capability

You can’t outsource your entire AI strategy. You need internal people who understand what these tools can and can’t do.

I’m not saying every IT team needs machine learning engineers. But you need someone who can evaluate vendor claims, design good tests, and think critically about where AI fits.

I usually recommend starting with one or two people who get dedicated time to learn and experiment. They become your internal advocates and bullshit detectors. Because there’s a lot of bullshit in the AI space right now.

When specialists in this space come in to help, they should be building your capability, not just doing the work for you. If your vendor isn’t transferring knowledge, find a different vendor.

Handle the People Issues

Technology’s usually the easy part. People are hard.

Every AI implementation changes someone’s job. Sometimes that’s threatening. Sometimes people assume they’re being replaced. Sometimes they’re just skeptical because they’ve seen other “transformation” projects fail.

I spend as much time on change management as I do on technology evaluation. That means clear communication about what’s changing and why. It means involving the people who do the work in designing the solution. It means being honest about what jobs might change.

Last year I worked with a client rolling out AI for customer service. We brought the service team in early. They identified which queries were repetitive (good for AI) and which needed human judgment (keep human). They designed the handoff process. They felt ownership, not threatened.

The project succeeded because the people who used it every day helped build it.

Set Realistic Expectations

AI won’t transform your business overnight. Anyone promising that is selling you something.

What AI can do is make specific processes faster, cheaper, or more accurate. Sometimes significantly so. But it requires ongoing management, monitoring, and adjustment.

I tell clients to think about AI like any other enterprise system. It needs governance, it needs maintenance, it needs people who understand it. Budget for that. Plan for that.

The wins come from sustained, focused application. Not from one big bang deployment.

What I’d Do First

If I were starting an AI adoption strategy today, I’d do this:

Spend one month identifying painful, repetitive processes. Talk to people doing the work. Document what takes time and why.

Pick the three most promising candidates. Run small, time-boxed tests. Measure everything. Kill the ones that don’t work. Double down on the ones that do.

Build internal knowledge as you go. Make someone responsible for AI capability building. Give them time and resources.

And be patient. Good AI adoption takes a year or two, not a quarter. But the results compound if you do it right.

That’s how you build an AI strategy that actually delivers value instead of just checking a box.